Half of unregulated AI interactions end with AI completing student work

Nearly half of all conversations between students and an unregulated artificial intelligence tool end with the AI completing the task for the student, according to data from the Danish AI company Miira. The company has deployed its AI on more than 40 primary schools, upper secondary schools and commercial colleges across the country. On roughly a quarter of those schools, no rules were set for how students could use the AI.

Miira’s data shows that in 49 percent of interactions with the unregulated version, the student asked the AI to produce a finished assignment, such as “write my analysis” or “write this as a finished submission”. Only 30 percent of students used the AI constructively, either to get learning help or to develop something together with the tool.

“It’s a disaster for learning,” said Miira’s director Laurits Rasmussen. “But it lies in the design of an unregulated AI. It is supposed to make life as easy as possible for the user, in the short term.”

The company has also developed a regulated “tutor AI” that never gives students the answer. In schools that used that version, the share of students who used the AI for learning or improvement rose to 47 percent.

Lillian Buus, head of the Research Centre for Didactics and Pedagogy at Via University College, called the figures alarming. “There is an acute need to teach children and young people how AI is used and how it is not used,” she said. Buus added that teachers face a major task in ensuring students develop skills orally as well.

On Tuesday, the association Danske Gymnasier called for an urgent change to the annual grading system, proposing that students no longer receive separate written and oral grades but a combined grade instead, as AI has made written work difficult to assess.